Pump Power Optimization of Ultra-wideband C+L-band Raman Amplifier Based on Neural Network Gain Prediction Model and PSO Algorithm

We propose a neural network-based gain prediction model for a 12 THz C+L-band distributed Raman amplifier, achieving a MSE of 0.029 and an R² value of 0.997. Using this pre-trained model as a surrogate simulator, particle swarm optimization efficiently determines the optimal six-wavelength pump power configuration, resulting in an average gain of 12.2 dB with a gain flatness of 2.58 dB across the entire C+L-band. When the optimized pump configuration is validated through simulation, the actual gain flatness differs by only 0.1 dB from our neural network prediction, demonstrating the high accuracy of our optimization framework for practical deployment in ultra-wideband optical systems.